Interpretable neural network forecasting of CO₂ emissions from renewable and non-renewable energy use in Saudi Arabia

Abstract

Energy-intensive economies need to precisely forecast CO₂ emissions associated with energy use to be well-prepared for sustainability. This study aims to develop a precise predictive framework for modeling Saudi Arabia’s CO 2 emissions using artificial neural networks (ANNs). A comprehensive national dataset covering both renewable and non-renewable energy consumption, as well as key environmental indicators for several years, was compiled and standardized. To train and test the proposed ANN model, a structured train-validation-test protocol was used. Likewise, the proposed model was compared to standard baseline models like linear regression and support vector regression. The findings show that the proposed ANN model consistently outperforms the baseline techniques, exhibiting reduced prediction errors and increased explanatory power across all evaluation criteria. Additionally, SHapley Additive exPlanations (SHAP) were utilized to quantify the contribution of each input variable to the model’s output to increase transparency. This allowed for both local and global understanding of the results. Thus, the study provides an interpretable neural network framework for energy-environment modeling that combines explainability and accurate prediction. By identifying the main causes of emissions and enabling data-driven supervision of national decarbonization plans, the findings provide useful information for energy policymakers and sustainability planners.